#83: The Physical AI Narrative is Gaining Steam

Robotics companies need specialized, real-world datasets

“Make sure to zoom in on how I scrub the grease stain!”

We are living in weird times. DoorDash is paying people to film themselves washing dishes.

Yes, you read that correctly. Last week, Bloomberg reported that DoorDash created an app for their “Taskers” to record themselves completing chores around the house, like dishwashing or folding clothes.

Why? It’s all about offline data collection.

As foundational model companies continually look to upgrade their AI, they’ll need new, niche datasets to maintain their edge. And believe it or not, there’s only so much data they can train on that is available online. Thus, they must look in a place no tech bro would have considered. The real, physical world.

Now, there’s plenty of niche datasets out there. So why is DoorDash offering compensation to those willing to complete everyday tasks on camera? Check out the quote below from the Bloomberg article.

“As an example of how the paid video submissions work, instructions for a dishwashing task ask that the person capture footage with a body-worn camera pointed down toward their hands, scrubbing and rinsing at least five dishes and holding each clean dish steady in frame for a few seconds before moving to the next dish. That camera footage may be valuable as robotics firms hone their humanoids’ ability to recognize objects.”

Companies building physical AI, ranging from autonomous vehicles to at-home robots, need to understand all the intricacies and variations of how people complete these everyday tasks to build a mechanical substitute.

It’s a bit dystopian. The gig economy workers may benefit in the short term by providing training data to DoorDash. But they’re training their robot replacements, as DoorDash may sell this data (they already have a partnership with Waymo) to robotics startups.

I believe 2026 is the year when the physical AI narrative starts to gain steam. Waymo has already completed over 20 million rides and is planning to launch in an additional 20 cities by year’s end. Also, you have Elon Musk shifting the storyline of Tesla from an electric vehicle manufacturer to a robotics company. Oh, and let’s not forget that the main highlight of NVIDIA’s 2026 GTC conference (some refer to it as the “Super Bowl for AI”) was innovation in physical AI that their semiconductor chips will enable.

Robotics outside the factory floor have always seemed far in the future. More so reserved for the plot of a sci-fi thriller. But as I went for a walk today in Manhattan, I strolled by a Waymo car in training. A person was in the car, but still, if an autonomous vehicle can figure out how to navigate traffic in New York City, ride-share drivers will need to find a different way to earn a living.

The world is changing. Candidly, I believe it will become quite normal to see someone filming themselves throughout the day because they are getting compensated by an AI company for doing it. AI valuations are through the roof, particularly on the private market side. Tech companies need to justify their valuations, and physical AI presents another avenue to live up to the hype. Yet, the clearest story is how robotics are likely to reduce the labor force. That’s a topic for another article.

Because tech firms need to live up to their current valuations, the demand for real-world data will continue to rise. The issue is on the supply side. It’s not easy to capture niche tasks at scale.

Which is why I believe that tech-enabled marketplaces employing the gig economy stand to reap significant rewards in the coming years as robotics companies look for training data.

In addition to delivery companies (i.e. DoorDash), marketplaces like Angi (formerly Angie’s List), Uber, and TaskRabbit are sitting on a treasure trove of data waiting to be unlocked. Looking to create a robot that mounts televisions? You’re going to want to strap a camera on a TaskRabbit TV mounting specialist. Or have a contractor on Angi film themself cleaning a pool and checking the chlorine levels.

Private equity is famous for rolling up* service firms like plumbers, electricians, and HVAC companies for the sake of streamlining processes and adding modern technology across the group. It’s become such a popular investment strategy that there are plenty of memes about the thesis. The joke here is that, believe it or not, running a real business isn’t as easy as the math makes it out to be in a spreadsheet. But perhaps this business idea has a new runway. There’s opportunity to roll up a business within a particular trade, record how each employee completes the task at hand, and then sell that data to a robotics startup.

For example, try landscaping companies. They complete both routine (lawnmowing, fertilization) and nuanced tasks like tree cutting, irrigation system installations, and patio building. I imagine within the next five years, we’ll see the first widely adopted autonomous lawnmower. But how could one build an autonomous lawnmower? They first must acquire a ton of specialized data on how people cut lawns today.

*By rolling up, I am referring to purchasing several similar companies. The idea is that by operating, say, 20 electrician businesses, there are operational efficiencies to be had with scale and those will lead to more effective operations.

Fast forward. Thanks to people willing to film themselves folding laundry, robots are here and ready to complete your chores. But do people really want to delegate their daily tasks to a non-human object?

It’ll be interesting to see how consumer adoption of physical AI plays out. I believe consumer adoption will lag, as people are scared by the narrative that Hollywood has portrayed about robots for decades. A robot living in your home and washing the dishes is likely many years past what any tech guru predicts. But then again, Waymo has done an excellent job with branding around autonomous driving. They claim that their cars are involved in “92% fewer crashes causing serious injuries or worse than human drivers.” Perhaps people will warm up to the robot family member sooner than I think.

But in the short-to-medium term, the opportunity for investing or making some extra cash is in the acquisition of offline data. If you told me I could get paid to do the dishes by DoorDash, if I filmed myself doing it, I would have been happy to clear the table as a kid. Sorry about that mom!